On the neural basis of image schemas Ellen Dodge and George Lakoff 1. Introduction The idea of image schemas emerged from the empirical research on spatial relations terms by Talmy (1972, 1975, 1978, 1983) and Langacker (1976, 1986) in the mid­1970's. They found, independently, that (1) even closely related languages vary widely in the meanings of their spatial relations terms, and that (2) despite this variation, the cross­linguistic differences could be analyzed in terms of combinations of universal schemas ­­ paths, bounded regions, contact, forces of various kinds, and so on ­­ together with metaphorical versions of these. Research since then has confirmed and extended their findings, with analyses done of languages around the world. The methodology behind these discoveries is commonplace within linguistics. (1) A cross­linguistic search within some natural domain (e.g., spatial relations) is conducted, uncovering diverse and complex systems. (2) A hypothesis is made that the complexity and diversity can be explained in terms of combinations of simple universal primitives. (3) Language­by­ language analysis is performed, showing that the same universal primitives combine differently in different languages to yield the observed complexity and diversity. Not all linguists work with this explicit methodology. The Nijmegen group, for instance, starts with the same first step of performing a cross­ linguistic search in a particular domain. In the domain of spatial relations, for example, the group uses a methodology based on line drawings (see Bowerman 1996; Bowerman and Choi 2001; Levinson et al 2003). Each drawing represents a spatial relation between two objects. The native consultant is to supply the term most appropriate for describing that relation. They find the first result that Talmy and Langacker found: there is a lot of diversity across languages regarding which words name which collection of pictures. 2 Ellen Dodge & George Lakoff Assuming that universal primitives would be relatively unitary concepts associated with individual words, they consequently argue against the presence of universal primitive and innate concepts. It is important to note, however, that these conclusions are based on a different notion of “universal primitive” than the one used by Talmy and Langacker, who assume that words may express concepts which are themselves complex combinations of primitives. Different assumptions about the nature of universal primitives may thus lead to different choices of methodology and to different conclusions. Since we take the Talmy­Langacker analyses as deep and insightful, and since we are interested in maximizing explanations from neuroscience, our results are, not surprisingly, very different from those of the Nijmegen group. By the early 1980's, image schema research had led to a deep question: Where do the universal primitive image schemas come from? Since then, two answers have been proposed: (1) Johnson (1987) saw them as arising from recurrent everyday bodily experiences, such as the early childhood experience of putting things into containers and taking them out again. (2) Regier (1995, 1996) argued that the human brain is structured naturally so as to compute the primitive universal image schemas, and to combine them. These are, of course, not mutually exclusive. Regier's hypothesis explains the types of experiences that Johnson observes. That is where things stand today. To get a better idea of the details, let us begin with Johnson's ideas. 1.1. Motion Experience Consider motion­related experiences. From the time you first learn to crawl, moving yourself around in the world is a significant part of your daily routine. On a typical morning you might get out of bed, leave the bedroom, and walk to the kitchen to get something to eat. Going outside, you may then jog around the block, run away from an angry dog, bicycle to work, or drive to the store. What types of image­schematic structures are associated with such basic experiences as these? To answer this, we first have to have some idea of what an image schema is. Johnson (1987), using an experiential approach, proposed the following defining elements: (1) recurrence across many different experiences; (2) a relatively small number of parts or components; and (3) an internal structure that supports 3 On the neural Basis of Image Schemas inferences. Considering the motion experiences just mentioned, we can see that while they clearly differ in some respects (e.g. what time of day it is, how far we move, whether food is involved in the experience), all of them have at least one very schematic commonality: they involve a change in the mover’s location. These locational changes have the structure of the very basic SOURCE­PATH­GOAL image schema (Johnson 1987; Lakoff 1987): we are one place when we start to move (a Source location); over time we change location (a series of Path locations); and when we stop moving, we are usually somewhere different than where we started (a Goal location). This very basic schema has few parts (Mover, Source, Path, Goal), applies over a very wide range of motion situations, and supports inferences (e.g. if you are at the goal location you have already been at the source and path locations). Thus, we see that recurrent, everyday motion experiences display at least one kind of image­schematic structure. 1.2. Moving Beyond Experience To gain a fuller understanding of image schemas, though, we need more than an experience­based analysis. Our next step will be to examine the image schemas associated with descriptions of motion events, using the linguistic methodology described in the introduction. We will start with an analysis of some English sentence examples, and will then look at work done on cross­linguistic variation and universality. An examination of motion descriptions indicates that, in addition to containing a variety of spatial image schemas, they contain other kinds of image schemas as well. Linguistic analysis thus helps us identify the existence of primitive image schemas. Moreover, linguistic analysis can lead to a better understanding of how different languages may use and combine these primitive image schemas. However, linguistic analysis does not, by itself, explain the origins of image schemas. And, even though image schemas may be associated with recurrent regularities in experience, an experiential analysis by itself does not explain how languages may make use of these schemas, nor does it explain why we perceive the particular schematic structures we do. In order to provide more complete explanations, we believe it is necessary to consider the role of the human brain. In the second half of the paper we will therefore pursue Regier’s line of argument that the brain computes image schemas. First, we review some of the work that supports the link 4 Ellen Dodge & George Lakoff between image schemas and neural structure. Then, we will once again focus on motion experiences and descriptions, exploring a hypothesis about the relation between image schemas and particular neural structures. Through such an analysis, we will show how the study of the brain not only can lead to a fuller understanding of image schemas, but also can provide insight into the relations between image schemas, experience, language, and the brain. 2. Image­schematic Structure in Language 2.1. Image Schemas, Experience, and Language Image schemas structure our experience independently of language. For example, we experience many things as containers — boxes, cups, baskets, our mouths, rooms, and so on. Prior to learning language, children go through a stage of exploration in which they repeatedly put things in and take them out of many different kinds of objects, thus treating these objects as containers. Starting at an even earlier age, they observe actions such as these being performed by others. As Mandler (this volume) describes, infants show some understanding of containment well before their first birthday. Image schemas play a vital role in fitting language to experience. Image schemas define classes of experiences that are characterized by the same word (e.g., in or out or up). This fact raises two questions: How is it possible for different experiences to have the same image­schematic structure? And how are image schemas expressed in language? We will start with the second question, and get to the first later in this essay. What is of note is that the way image schemas are expressed in language is a central feature of linguistic structure. Image schemas may be expressed through the use of many different linguistic forms, including prepositions, postpositions, verbs, cases, body­part metaphors, or morphemes (e.g., in Cora, as described by Casad and Langacker (1975)). Which forms a given language uses is a typological feature of that language. 2.2. English examples 5 On the neural Basis of Image Schemas When we describe a person’s motion, what sorts of image­schematic structure do we express? Consider the following two sentences: 1) He walked to the kitchen; and 2) He walked into the kitchen. Both of these sentences indicate that a mover (he) is changing location with respect to a “landmark” (the kitchen). But these locational changes are only schematically specified; all we can really infer from these sentences is that initially the mover was not at the kitchen, then he moved, and after moving the mover was at or in the kitchen. Both of these motion descriptions thus seem to express SOURCE­PATH­GOAL schematic structure. The second sentence expresses additional schematic structure, which serves to further specify the spatial relation between the mover and the landmark. The first sentence only indicates, roughly, that at the end of motion the mover is somehow co­located with the kitchen. This may mean that the mover is inside the kitchen, but could also mean that he is directly outside it, or that he is standing in the doorway.